Crystal: A scalable and fault-tolerant Archimedean-based server-centric cloud data center network architecture

The explosive increase in adopting cloud services leads to the need for accommodating millions to even billions of servers in the supporting infrastructure of data center networks. Thus, providing highly scalable and noticeably fault-tolerant interconnecting architectures with great network capacity, limited delay, and reasonable energy consumption has become of paramount importance. In this paper, Crystal, a server-centric and recursively constructed structure, is presented. Crystal is made out of building blocks inspired by vertex configuration pattern of Archimedean solids and is intrinsically fault-tolerant mainly due to the adoption of multihoming mechanism in the building block. Crystal also scales double exponentially owing to its recursively defined architecture. Furthermore, to increase the routing efficiency and manage different kinds of failures in the network, two exclusively designed routing algorithms are proposed. Theoretical analysis and simulation results all witness that Crystal is highly fault-tolerant and can provide good network capacity and low latency to support delay-sensitive and data-intensive applications. Generally, Crystal satisfies a reasonable trade-off between all the major goals of designing a cloud data center network.